How much reductionism?
The question of being pro- or anti-reductionist came up briefly in a recent lab meeting. This is a re-hash of a piece I wrote a few years ago in response to a research funding allocation question that touches on that subject. It relates to a question that was being posed by the government Agriculture/Environment department which supported much of the work I did back then. The specific example is in the context of designing a science program to address a policy question, but I think the method may be useful at the start of the design process for any new program of research.
At one point in the policy-science-policy cycle we (the scientists) were asked to design a program of work to investigate the sustainability of crop farming. The policy makers/science funders also asked us to form a large collaborative team from the scientists at two different institutes to design the program and carry out the work. We were allocated an overall budget (which was simply the sum of the two existing programs of research). The two teams met on several occasions and did a reasonable job of combining their different approaches and separate organizantional goals, except for one major component of the program. One team, let's call them the blue team, wanted to do on-farm research with commercial farmers and examine questions of sustainability by collecting extensive, but possibly imprecise, data across many examples. The red team wanted to set up a controlled, replicated systems experiment in a single location and study detailed ecological and biophysical processes. There were not enough resources to do both. So, the policy makers had a classic economic problem of allocating scarce resources in the best way possible to achieve a goal - in this case to develop methods for studying sustainability in agriculture. The teams reached a standoff and reported back to the funders that they should simply pick one of the two proposed approaches and everyone would fall in line behind that choice. On the one hand we had a program of work that addressed the issue of farm sustainabilty at the scale of the farm and the crops each farm produced but was essentially observational and had no true replication, and on the other hand a program of work looking at processes in individual crops at the sacle of individal organisms with lots of replication and control of many exogenous sources of variation. The funders asked for some further guidance as to how to make the decision between the two.
Within my small group of modelers and analysts (part of the blue team) we decided to answer the request by turning some of the analytical and theoretical apparatus we used for studying systems on ourselves. The result was a simple procedure for deciding what the appropriate organizational scale will be for addressing any research question (and hence, in the particular case in question, which research to fund). The approach pulled together ideas gathered from Arthur Koestler's pioneering work on hierarchicial systems "The Ghost in the Machine", and Pete Turchin's monograph "Complex Population Dynamics", together with work we were doing looking at methods for modeling policy evolution. Our procedure went by the snappy(ish) nemonic IOUORMI (I owe you or me, which conveys a vague sense of a zero sum game or inherent trade off in which I'll either have to concede something to my opponent or to myself). The 7 letters stand for the following stages/approaches
Identify the Objects - work out what it is that you want to know about (in the example in question we were interested in the sustainability of farms).
Use Occam's Razor - don't elaborate your explanation or your explanatory apparatus beyond what is needed to successfully answer your question.
try Methodological Individualism - having located the objects of interest in their natural organizational hierarchy, one way to come up with simple approaches to answering questions about their behavior is to look at processes operating the next scale down in the hierarchy, together with behavior of the objects at scale of interest. It doesn't always work, but it's probably a good starting point in the absence of other design principles. Our contention was that this approach would be likely to identify the appropriate scales at which research should be conducted, allowing policy makers to choose between the competing options being offered.
In the end we won the argument but the decision went the other way for politiical reasons that were outside of the terms of the discussion - a briefly painful but useful lesson about science-policy interaction in itself.